Background Documents on

Risk Assessment in Engineering

Document #5

Optimization with a LQI Acceptance
Criterion




JCSS
Joint Committee of Structural Safety


Date: November 2008
JCSS author: Rdiger Rackwitz, Technical University
Munich

Optimization with a LQI Acceptance Criterion
H. Streicher and R. Rackwitz
Draft, February 2006
1. Introduction
This example collection covers some typical applications in structural reliability and discusses
several important aspects. It is meant to be the basis by example applications for practicable accep-
tance criteria for structural codes. The first three examples show the influence of the interest rates
in an optimization for the public or the owner, the optimal and acceptable solutions for different
cost values and the importance of the coefficient of variation of the resistance and load variables.
Especially in the first two examples a number of parameter studies are performed. The next exam-
ple deals with the realistic design for the buckling of a reinforced concrete column. Two examples
for the optimal design under reliability constraints for different load combinations of intermittent
load processes will be evaluated. Finally, an example from earthquake engineering is included.
The obective functions are based on a systematic recontruction policy and a constant benefit rate.
Optimization is carried out by the methods described in [9] but other methods are also possible.
All LQI considerations are based on [6] . Usually, the LQI-criterion is added to the cost-benefit
optimization task as a constraint. In [6] it is shown that the LQI-criterion requires
dC
Y
(p) G
x
kN
PE
dh(p) (1)
with p a vector of design parameters. Using dF(x) = f(x)dx one again solves this equation as an
optimization task
Minimize: S(p) = C(p) +G
x
kN
F
h(p) (2)
because eq. (1) is the first-order optimality condition of eq. (2) if no cost-benefit optimization is
performed.
From [6] one concludes that the societal value of a statistical life is G
x
=
g
q
C
x
where x
stands for the particular mortality reduction scheme. For constant mortality reductions , a Eu-
ropean population with GDP 25000 PPPUS$(year 2000), the part available for risk reduction
g 17500 PPPUS$ and q 0.15, C 45 and G 5Mill US$ if no discounting and no age-
averaging. If discounting and age-averaging is performed it is C
d
16 and G
d
1.9Mill US$
if discounting and age-averaging is performed and constant mortality reductions are envisaged by
a safety measure. These values then enable to compute the societal willingness-to-pay to save a
statistical life by SWTP =
g
q
C
x
dm.
The examples have taken from several sources but many of them are contained in a collection
of examples in [2] . Therefore, some variations in important parameters, especially for the LQI-
acceptability criterion, remain.
1
2. Example 1: Resistance-demand problem
The example has already been given in [5] and in [7] in somewhat different form and with differ-
ent parameters. A single-mode system is considered where failure is defined if a random resistance
or capacity is exceeded by a random demand. The demand is modelled as a one-dimensional,
stationary marked Poissonian pulse process of disturbances (earthquakes, wind storms, explo-
sions, etc.) with stationary renewal rate and random, independent sizes of the disturbances
S
i
, i = 1, 2, ..... The disturbances are assumed to be short as compared to their mean interarrival
times. We study normally distributed and log-normally resistance and disturbances. The resistance
has mean p and a coefficient of variation V
R
. The disturbances are independent and have mean
equal to unity and coefficient of variation V
S
so that p can be interpreted as central safety factor.
p is taken as the only optimization parameter. For failure once a disturbance occurs with normally
distributed variables we have:
P
f
(p) =

p 1
p
(pV
R
)
2
+V
2
S
!
(3)
In this model both resistances and disturbances can have negative values which usually is incor-
rect from a physical point of view. Therefore, we truncate the distributions at zero and a more
complicated formula for the failure probability results:
P
f
(p) =
1
(
1
V
S
)
1
(
1
V
R
)
Z

0
1

2V
S
(
s 1
V
S
)
Z
s
0
1

2V
R
p
(
r p
V
R
p
)drds (4)
For the log-normal case we have:
P
f
(p) =

ln
n
p
q
1+V
2
S
1+V
2
R
o
p
ln((1 +V
2
R
)(1 +V
2
S
))

(5)
An appropriate objective function for systematic reconstruction which will be maximized, is then
given by
Z(p) =
b

C(p) (C(p) +H
M
+H
F
)
P
f
(p)

(6)
where C(p) = C
0
+C
1
p
a
and H
M
the initial material cost.
The acceptability criterion can be written as
d
dp
C(p) G

kN
PE
d
dp
P
f
(p) (7)
The parameter assumptions are: C
0
= 10
6
, C
1
= 10
4
, a = 1.25, H
M
= 3 C
0
, V
R
= 0.2, V
S
= 0.3
and = 1 [1/year]. The LQI-data are e = 77, g = 25000, C

= 25, q = 0.16, kN
PE
= 10 so
that H
F
7 10
6
and G

(, ) 5 10
6
. Monetary values are in US$. Optimization will first be
performed for the public. Therefore, compensation cost are included in the objective function. The
benefit rate is b = 0.02 C
0
and the interest rate is = 0.0185. For the failure probability model eq.
(3) one finds the optimum for p

= 5.54 and a limiting value of p

lim
= 3.86. If one uses the model
in eq. (4) the optimum is at p

= 3.00 and the limiting criterion eq. (7) results in p

lim
= 2.49.
Finally, for the model in eq. (5) the optimum is at p

= 4.40 and has a limit of p

lim
= 3.45. The
three objective functions are shown in figure 1. One concludes from this study that the stochastic
models for the failure function must be as realistic as possible. All subsequent investigations are
2
based on log-normal distributions.
0 2 4 6 8 10
1
0.5
0
0.5
1
p
Z
(
p
)
plim(nt)
plim(ln)
plim(n)
Figure 1: Objective functions and limiting criteria for the failure models in eq. (1) (dashed line),
(2) (solid line) and (3) (dotted line)
The influence of the cost C
1
on the optimal and acceptable solution is shown in figure 2, where
C
1
varies from 1000.0 to 100000.0 and the corresponding optimal and acceptable failure rates are
given. For C
1
> 12000 the objective function Z(p

) is negative.
1
.
10
3
1
.
10
4
1
.
10
5
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
as :=
L
o
g

(
F
a
i
l
u
r
e

R
a
t
e
)
Log (C
1
)
1
.
10
3
1
.
10
4
1
.
10
5
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
as :=
L
o
g

(
F
a
i
l
u
r
e

R
a
t
e
)
Log (C
1
)
Figure 2: Failure rates for various cost C
1
. Dashed lines correspond to acceptable, solid lines to
optimal results.
The optimal solution is acceptable for all cost values C
1
. It is most important to ensure that all
necessary investments into life saving in order to guarantee the safety for human life and limb are
done and the resulting failure rate of a technical facility is acceptable . This is already fulfilled if
the acceptability criterion derived from the LQI is fulfilled.
The owner uses some typical value b = 0.07 C
0
. Compensation cost (Life saving cost) are
not included in the optimization. The objective of the owner is in general to maximize the benefit
3
from an investment. Therefore, he uses a higher interest rate =
owner
given by the financial
market. But he is still bound to build safe structures. This is only guaranteed if the LQI-criterion
which may includes the ratio /
public
of the economic and the public interest rate
public
given by
the characteristics of a society is fulfilled. The corresponding failure rates of the optimization for
various interest rates
owner
are given in figure 3 with kN
F
= 50. In general for a long period
the ratio /
public
is or is close to 1. For shorter service times or credit periods a different value is
possible. Therefore, the acceptable failure rate is shown in figure 8 for /
public
= 0.7, /
public
=
1.0 and /
public
= 1.3.
0 0.02 0.04 0.06
1
.
10
5
1
.
10
4
a :=
L
o
g

(
F
a
i
l
u
r
e

R
a
t
e
)
Owners interest rate
/
public
= 0.7
/
public
= 1.0
/
public
= 1.3
0 0.02 0.04 0.06
1
.
10
5
1
.
10
4
a :=
L
o
g

(
F
a
i
l
u
r
e

R
a
t
e
)
Owners interest rate
/
public
= 0.7
/
public
= 1.0
/
public
= 1.3
Figure 3: Failure rates for various interest rates of the owner. Dashed lines correspond to
acceptable, solid lines to optimal results.
It can be seen, that the LQI-criterion (7) becomes active for increasing interest rates
owner
de-
pending on the ratio /
public
. In that case the owner must use the acceptable solution. Further
calculations show, that for kN
F
> 50 the LQI-criterion becomes permanently active. kN
F
< 50
makes the optimal solution always acceptable.
For the standard case the ratio H/C
0
is varied between 1 and 10 in figure 4, a range which covers
most applications. It is seen that increasing H by an order of magnitude decreases the optimal
failure rate by roughly half an order of magnitude.
0 2 4 6 8 10
6
.
10
6
8
.
10
6
1
.
10
5
1.2
.
10
5
1.4
.
10
5
H/C0
O
p
t
r
a
t
e
H/C0
Figure 4: Optimal failure rate with increasing ratio H/C
0
This example also allows to derive risk-consequence curves by varying the number of fatalities
in an event. With the same data as before but SLSC = 7 10
5
and G
x
(, ) = 4 10
6
for N
F
= 1
4
we first vary the cost effectiveness of the safety measure (see figure 5). Here, only the ratio C
1
/C
0
is changed. The upper bounds (solid lines) are derived from eq. (7) and the lower bounds (dashed
lines) corresponds to the societal optimum according to eq. (6) (b
S
= 0.02C
0
,
S
= 0.0185).
1 10 100 1
.
10
3 1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
0.1
NF
h
C1/C0 = 0.1
C1/C0 = 0.01
C1/C0 = 0.001
Figure 5: Acceptable failure rate over number of fatalities for different C
1
/C
0
for first example.
Dashed lines correspond to optimal solution for the public.
Most realistic is probably a ratio of C
1
/C
0
= 0.001. The failure rate of approximately 10
4
per
year for N
F
= 1 corresponds well with the controllable crude mortality of the same magnitude.
Some more discussion is provided in the next example.
3. Example 2: Resistance-demand problem - variable discounting
The same failure model as in example 1 (eq. (5)) is used with variable discounting. An appro-
priate objective function then is:
Z(p) =
b
C
0
Z

0
e

R
t
0
()d
dt

1 +
C
1
C
0
p
a

1 +
C
1
C
0
p
a
+
H
M
C
0
+
H
F
C
0
Z

0
e

R
t
0
()d
g(t, p)dt (8)
with
g(t, p) =

X
m=1
P
f
(p)f
m
(t)(1 P
f
(p))
m1
(9)
f
m
(t) =
(t)
m1
(m)
exp[t] (10)
5
and
(t) = +
max
exp(at); = 0.02, a = 0.013 (11)
Here, we use the exact computation formula for g(t, p) and not the approximation proposed in
[7] . In eq. (8) the first term is the benefit derived from the existence of the facility. A constant
benefit rate b is assumed and a time-depending discount rate (t). The second term is the erection
cost where C
0
is the constant part and C
1
p
a
the part depending on the design parameter p. The
third term are the damage cost. Because systematic reconstruction is chosen the cost term includes
reconstruction cost, cost H
M
due to physical damage, and the compensation cost for human lives
H
F
. Reconstruction times are assumed negligibly small. The details are discussed in [7] .
The acceptability criterion has the form:
d
dp
(C
0
+C
1
p
a
) G

`
kN
PE
d
dp
(P
f
(p)) (12)
Some more or less realistic, typical parameter assumptions are: C
0
= 10
6
, C
1
= 10
4
, a = 1.25,
H
M
= 3 C
0
, V
R
= 0.2, V
S
= 0.3, b = 0.05C
0
and = 1 [1/year] . A constant benefit rate of
b = 0.05C
0
is chosen. The LQI-data is e = 77, GDP = 25000, g = 17500, q = 0.15. Also
there is N
PE
= 100, k = 0.1 so that H
F
= SLSC kN
PE
= 5.4 10
6
and G

kN
PE
= 5.0 10
7
,
respectively. The value of N
PE
is chosen relatively large for demonstration purposes. Also, the
values of H
F
and G

kN
PE
are conservatively chosen. Monetary values are in US$.
From eq. (8) one concludes that there is a maximum discount rate in order to render it positive.
Projects whose objective is negative at the optimumdo not make sense. Keeping = 0.02 constant
one can determine the maximum time preference rate from

max
= solution of (Z(p
opt
, ) = 0) (13)
which is
max
= 0.035.
Performing the optimization with the (maximum) discount rate
max
gives p
opt
= 4.04 (r(p
opt
) =
3.0 10
5
) leaving the objective function positive but close to zero. Criterion (12) requires p
lim
=
3.61 (r(p
lim
) = 1.1 10
4
). It is interesting to see that in this case one can do better in adopting
the optimal solution rather than just realizing the facility at its acceptability limit (see fig. 6). If,
however, compensation cost are omitted from eq. (8) then criterion (12) is situated very close to
the optimum. Increasing the discount rate moves the optimum insignificantly to the left.
2 4 6 8 10
0.4
0.2
0
0.2
0.4
p
Z
(
p
)
p
lim
= 0.03
= 0.035
Figure 6: Objective function and acceptability limit for capacity-demand example for two
6
discount rates (solid line: with compensation cost, dashed line: without compensation cost)
For smaller (or larger benefit rate) the maximum of the objective function (with compensation
cost) is well above zero as shown in fig. 6 (dotted line). The location of the optimum varies very
little with the discount rate as shown in fig. 7.
0 0.01 0.02 0.03 0.04
1
.
10
5
1
.
10
4
rho
o
p
t

r
a
t
e
a
Figure 7: Location of optimum as a function of
The stochastic model and the variability of capacity and demand also play an important role for
the magnitude and location of the optimum as well as on the position of the acceptability limit.
The specific marginal cost (rate of change) of a safety measure and its effect on a reduction of the
failure rate are equally important as pointed out already in [5] . This is shown in fig. 8 where the
acceptable failure rates are plotted over the parameters C
1
and V
R
for G

`
kN
PE
= 5.0 10
6
,i.e.
kN
PE
= 1, and C
0
= 10
6
.
1
.
10
3
1
.
10
4
1
.
10
5
1
.
10
4
1
.
10
3
0.01
C1
rate
0 0.2 0.4 0.6
1
.
10
4
1
.
10
3
0.01
VR
rate
VR = 0.2
VS = 0.3
VS = 0.5
VS = 0.3
VS = 0.1
C1 = 10^4
Figure 8: Location of acceptable failure rate as function of marginal cost for safety and
coefficients of variation
This example also allows to vary the upper value m in eq. (9). It gives an impression about the
effect to consider only a limited number of reconstructions on the optimum. As seen in figure 9,
the effect can be remarkable.
7
2 2.5 3 3.5 4 4.5 5
0.15
0.1
0.05
0
p
Z
(
p
)
m = 1 2 3 5 10 50 100
Figure 9: Optimum solutions for different numbers of reconstructions
For this example risk-consequence curves can also be computed by varying the number of fa-
talities in an event. For the same data as before we first vary the cost effectiveness of the safety
measure. Only the ratio C
1
/C
0
is changed. The upper bounds (solid lines) are derived from eq.
(12). The lower bounds (dashed lines) correspond to the optimumaccording to eq. (8) (see left hand
side of fig. 10).
max
also must vary (decrease) with the failure consequences (number of fatalities
N
F
) for a given ratio C
1
/C
0
. This is taken into account on the right hand side of fig. 10 indicating
that the area between solid and dashed lines broadens for very high failure consequences.
1 10 100 1
.
10
3 1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
NF
r
C1/C0 = 0.1
C1/C0 = 0.01
C1/C0 = 0.001
1 10 100 1
.
10
3 1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
NF
r
C1/C0 = 0.1
C1/C0 = 0.01
C1/C0 = 0.001
1 10 100 1
.
10
3 1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
NF
r
C1/C0 = 0.1
C1/C0 = 0.01
C1/C0 = 0.001
1 10 100 1
.
10
3 1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
0.01
NF
r
C1/C0 = 0.1
C1/C0 = 0.01
C1/C0 = 0.001
Figure 10: Risk-Consequence curves for various ratios C
1
/C
0
(V
R
= 0.2, V
S
= 0.3,

max
= 0.035)
Again, normal structures have a ratio of around C
1
/C
0
= 0.001. A ratio of C
1
/C
0
= 0.01 or
higher may apply for earthquake resistant structures. Note that in these figures the failure rate is
8
given by P
f
and the number of fatalities is given by N
F
= kN
PE
. Therefore, these figures cover
the full range of and P
f
and k and N
PE
, respectively. The curves in fig. 10 are not classical F-N-
curves seen frequently in the literature where the exceedance probability instead of the failure rate
is plotted over the number of fatalities. The acceptable risk-consequence curves show an almost
perfect dependence of the type r
acc
(N
F
) =
K
N
F
with K = 10
3
(case with large variabilities
and/or costly safety measures), 10
4
(standard case) and, may be, 10
5
(for small variabilities and
inexpensive safety measures). Reasonable variations in the other parameters change these constants
at most by a factor of 2 to 4 as seen from fig. 11 where the extremes of the demographic constant
are shown. In particular, using G

`
= 1.9Mill PPPUS$ would move the curves in fig. 11 upwards
be a factor of 3.
1 10 100 1
.
10
3 1
.
10
7
1
.
10
6
1
.
10
5
1
.
10
4
1
.
10
3
NF
r
Cx = 15
Cx = 60
Figure 11: Risk consequence curve with two extreme of the demographic constant for
C
1
/C
0
= 0.001.
The demographic constant C
x
15 corresponds to discounted life expectancies with age-averaging.
C
x
60 corresponds to very old populations without any discounting and age-averaging. The
curves also indicate that larger GDPs lead to smaller acceptable risks and larger qs can diminish
them.
4. Example 3: Bending of a rectangular reinforced concrete beam
The benefit b derived from the structure and the interest rate vary with respect to the publics and
the owners interest. An appropriate objective function for systematic reconstruction and stationary
Poissonian disturbances is already given by eq. (6). C(p) = C
0
+ (400 [CU/m
3
] w d + 1000
[CU/m
3
] a
s
) 5 [m] are the construction cost with C
0
= 1000 and the reinforcement area a
s
[cm
2
],
width w [cm] and effective depth d [cm] of the cross section. Two constraints are defined on the
design parameter (a
s
, w, d): a maximum allowable reinforcement area a
s
4% w d and a lower
bound for the cross-sectional area w d 0.01 [m
2
]. The stochastic characteristics of the random
9
variables are given by
Variable Distr. Mean/
St.deviat.
Yield stress T
s

N
mm
2

N 360/36
Conc. strength T
c

N
mm
2

LN 40/6
Appl. bend. mom. M
b
[MNm] Gumbel 0.05/0.003
Model uncert. K [] Rect. 0.5/0.667
The limit state function is determined by the simplified model
g(x, p) =

1 K
a
s
T
s
w d T
c

a
s
d T
s
M
b
(14)
and the acceptability criterion by
5
p
C(p) G

`
kN
PE
5
p
(P
f
(p)) (15)
The LQI-data are e = 77, g = 25010, C
`
= 16, q = 0.16, kN
PE
= 0.1, = 1[1/year]
and therefore H
F
= 88000 and G

`
kN
PE
= 1.9 10
5
. The direct material damage cost are
H
M
= 3 C
0
= 3000. kN
F
= 0.1 < 1 takes account of the fact, that failure is ductile.
Optimization will first be performed for the public. The interest rate = 0.018 and the benefit
b = 0.02C
0
= 20 are chosen. The optimal solution is then (a

s
, w

, d

) = (14.45, 15.0, 24.09), the

corresponding failure rate 2.5110
7
and Z(p

) = 30.34. The LQI-criterion (15) is already fulfilled

for (a

s,lim
, w

lim
, d

lim
) = (13.41, 15.0, 22.37) with failure rate 1.5510
5
and Z(p

lim
) = 42.0. It is
interesting to note that for kN
F
= 1 the limiting failure rate is one order of magnitude higher. Here,
the optimization for the public gives also a more economic result than the acceptable solution and is
preferable. In this example the life saving cost (compensation cost) H
F
dominate the optimization
compared to the low cost H
M
and low marginal cost C
1
. Therefore, the solution is sensitive to the
value of the parameter kN
PE
. This is demostrated in figure 12. If a large number of fatalities is
expected the acceptable failure rate decreases. This is, however, not very likely because failure of
under-reinforced beams is ductile and associated with distinct warnings by large deflections.
0 0.01 0.02 0.03 0.04 0.05
1
.
10
9
1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
Interest rate
public
L
o
g
(
F
a
i
l
u
r
e
r
a
t
e
)
kN
F
= 0.01
kN
F
= 0.1
kN
F
= 1
0 0.01 0.02 0.03 0.04 0.05
1
.
10
9
1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
Interest rate
public
L
o
g
(
F
a
i
l
u
r
e
r
a
t
e
)
kN
F
= 0.01
kN
F
= 0.1
kN
F
= 1
Figure 12: Optimal and acceptable solutions for parameter values k N
F
. Dashed lines
correspond to acceptable, solid lines to optimal results.
For a maximum admissible interest rate ([4] )
max
= 0.0185 the objective function is about
Z(p

) 0 with (a

s
, w

, d

) = (14.44, 15.0, 24.07). If

max
is chosen the benefit derived from
a structure is balanced against all cost including initial, life saving and maintenance cost, which is
10
preferable from a societys point of view. The objective of the owner is in general to benefit from
an investment. Typical values are then b = 0.07 C
0
= 70 and = 0.05. The optimization yields
without including failure cost to (a

s
, w

, d

) = (13.37, 15.0, 22.29), a corresponding failure rate

of 1.86 10
5
and Z(p

) = 324.9. But the criterion (15) is not fulfilled. Therefore, the solution
is not acceptable. It requires at least (a

s,lim
, w

lim
, d

lim
) = (13.41, 15.0, 22.37) with failure rate
1.55 10
5
and Z(p

lim
) = 324.7.
It can be seen, that in this example the optimal solution for the public gives a more economic
result than the limit point derived from criterion (15) which is not active. The example is one
where it is comparatively cheap to improve safety.
5. Example 4: Slender steel column
The next example is a steel column with I-section and flange breadth
b
, flange thickness
d
and
height of steel profile
h
. The design parameter
d
and
h
will be changed in the cost optimization.
They are mean values of appropriate stochastic variables.
The objective function for systematic reconstruction and a stationary Poissonian renewal process
of disturbances is given in eq. (6). C(p) = C
0
+ C
1
(2
b

d
+ 10
h
) s 10
9
are the construction
cost of the steel column with length s = 9500 [mm], initial cost C
0
= 10000 [CU] and C
1
= 10000
[CU/m
3
]. = 10.1 [1/year] is a stationary renewal rate. The stochastic characteristics of the
random variables are given by
Stochastic variable Distr. Mean/St.deviat.
Yield stress F
S

N
mm
2

N 500/25
Load P
1
=P
2
[N] N 800000/150000
Load P
3
[N] LN 800000/200000
Web thickness t [mm] N 10/0.5
Flange Breadth B [mm] LN
b
= 300/3.0
Flange Thickness D [mm] LN
d
/2.0
Height of profile H [mm] LN
h
/5.0
Initial deflection F
0
[mm] N m
F
0
/
F
0
Youngs Modulus E

N
mm
2

N 210000/4200
where the mean value m
F
0
of F
0
is defined by
m
F
0
=
1
20
s
0.5 B D H
2
+
tH
3
12
2 B D +t H
+
s
500
and
F0
= 0.3 m
F0
. The limit state function for failure due to instability can be written approxi-
mately as
g(x, p) = F
S
P (
1
A
S
+
F
0
M
S


b

b
P
) (16)
where the auxiliary functions are defined by
P = P
1
+P
2
+P
3
A
S
= 2BD (area of section)
M
S
= BDH (modulus of section)
M
i
=
1
2
BDH
2
(moment of inertia)

b
=

2
EM
i
s
2
(Euler buckling load)
11
The acceptability criterion is given in eq. (15) and the LQI-data are e = 77, g = 25010, C
`
=
16, q = 0.16, kN
PE
= 1, and therefore H
F
= 8.8 10
5
and G

`
kN
PE
= 1.9 10
6
. The direct cost
are H
M
= 3 C
0
= 30000. The objective function Z(p) including the cost H
F
for the public with
b = 200 and = 0.0172 is shown in figure 13.
Z(p)

d
Z(p)

d
Figure 13: Objective function of the steel column.
An optimal curve can be found with respect to
d
and
h
(see figure 13). The LQI criterion is
not active in an optimization for the public. The corresponding failure rates are between 2.8 10
7
for
d
= 18.0,
h
= 421.9, Z
max
(p) = 186 and 1.9 10
7
for
d
= 22.0,
h
= 333.5, Z
max
(p) =
46.95.
18 19 20 21 22
250
300
350
400
450

d
Z
m
a
x
,p
u
b
lic
(p
)
Z(p)
Z
m
a
x
,c
w
n
e
r
(p
)
18 19 20 21 22
0
100
200
300

d
Z
max,public
(p)

h
18 19 20 21 22
250
300
350
400
450

d
Z
m
a
x
,p
u
b
lic
(p
)
Z(p)
Z
m
a
x
,c
w
n
e
r
(p
)
18 19 20 21 22
0
100
200
300

d
Z
max,public
(p)

h
Figure 14: Optimal objective functions for the steel column. Dashed lines correspond to
acceptable, solid lines to optimal results.
Additionally optimization for the owner is done in figure 14 with benefit b = 700 and interest
rate
owner
= 0.06. The curve is unacceptable compared with the solution derived from the LQI-
criterion (15) if
d
< 18.04 or
h
> 355.9. Then, the owner must observe the acceptability
criterion.
12
6. Example 5: Reinforced concrete column
This example is based on a detailed study in [3] . According to the design rules of Eurocode 1
and 2 the buckling of a reinforced concrete column of a multi-story building given in figure 15 is
considered.



L
a
1

a
2

n h
s

a
1
a
1

F
W
F
l
F
k



L
a
1

a
2

n h
s

a
1
a
1

F
W
F
l
F
k
Figure 15: Multi-storey building with loads.
The design parameter is the reinforcement area A
s
of the square shape b h of the column cross
section. A
s
should satisfy the condition 0.003 b h < A
s
< 0.08 b h. The limit state function is
expressed as the difference between the resistance bending moment M
R
and the actual bending
moment M about the cross section centre
G(x, p) =
R
M
R

E
M (17)
where the model uncertainties
R
and
E
are normally distributed with mean 1.1 and 1.0 respec-
tively and coefficient of variation V = 0.1.
The axial force N of the column is considered as the following sum of the single axial forces:
N = N
W
+N
imp
+N
Wind
where N
W
is the axial force due to self weight, N
imp
is the axial force due to imposed load and
N
Wind
is the axial force due to wind action (positive values are accepted for compressive forces).
The values are calculated as
N
W
= (n +1)a
1
a
2
t
c
/2
where
c
is the weight of the concrete per unit volume and n is the number of storeys with height
h
s
= 3 m above the considered column.
The imposed load from n storeys is
N
imp
= na
1
a
2
p
imp
/2
where p
imp
is given as sum of long term p
l
and short term p
s
imposed load.
N
Wind
is given as
N
Wind
= (
1
2
(L +nh
s
)
2
a
2
C
p
Gp
wind
4M
0
)/(3a
1
)
where L is the height of the column, C
p
the shape factor, G the gust factor and p
wind
the wind
pressure. The first order bending moment M
0
at the bottom of the column is assumed to be caused
13
only by wind action and is approximately given as
M
0
= L[C
p
Gp
wind
(L +nh
s
)a
2
] /8
The actual bending moment M is then determined by
M = M
0
+N(e
a
+e
2
)
where e
a
= L/2 is the additional eccentricity taking into account geometric imperfections with
initial sway . The second order eccentricity e
2
taking into account deformations of the column is
given as
e
2
= 0.1K
1
L
2
(2K
2
(F
s
/E)/(0.9(h d
1
)))
where E = 200000 N/mm
2
is the modulus of elasticity and d
1
the distance of the reinforcing bars
from the edge. The factor K
1
is calculated as
K
1
=

0, if 15

20
0.75, if 15 < < 35
1, if 35
with = L/h/3.4641.
K
2
is given by
0 K
2
= (N
u
N)/(N
u
N
bal
) 1
where the ultimate capacity N
u
of the cross section is
N
u
= bhF
c
+A
s
F
s
with as a reduction factor taking account of long term effects on the compressive strength, F
c
the
concrete strength and F
s
the yield strength.
The balanced axial force which maximizes the ultimate moment of the cross section is taken for
symmetrical reinforcement as
N
bal
= bhF
c
/2
The limit state function is then given using a simplified formula:
(A
s
F
s
(h d
1
d
2
)/2 +hN(0.5 N/(2hbF
c
)))
R

E
M > 0, if N < N
bal
K
2
(A
s
F
s
(h d
1
d
2
)/2 + bh
2
F
c
/8)
R

E
M > 0, if N N
bal
where d
1
, d
2
is the distance of the reinforcing bars from the edge.
This mechanical model is in various ways an approximation which, however, is fully sufficient
14
for the purpose. The characteristics of the stochastic variables are given in the following table
Stochastic variable Distr. Mean/St.deviat.
Concrete strength F
c

N
mm
2

LN 30/5
Yield strength F
S

N
mm
2

LN 560/30
Reduct. Factor Rect. 0.8/0.95
Width of section b[m] N b/0.005
Height of section h[m] N h/0.01
Distance of bars d
1
, d
2
[m] N 0.05/0.01
Initial Sway [rad] N 0/0.003
Model Uncert. load
E
[] N 1.0/0.1
Model Uncert. Resist.
R
[] N 1.1/0.11
Weight
c

MN
m
2

N 0.024/0.00192
Shape factor C
p
[] N 1.3/0.13
Gust factor G[] GUM 2.5/0.25
Adetailed description of all randomvariables can be found in [3] . The loads are modelled as rec-
tangular wave renewal processes with Gumbel distribution for the amplitudes. It is distinguished
between wind pressure with mean 0.00035 [MN/m
2
], standard deviation 0.00006 and jump rate
= 10 [year], long term imposed load with mean 0.0006 [MN/m
2
], standard deviation corre-
sponding to the random point-in time distribution and jump rate = 1/7 [1/year] and short term
imposed load with mean 0.0002 [MN/m
2
], standard deviation corresponding to the distribution of
24 hours maximum and jump rate = n [1/year] (n = number of storeys).
An objective function for systematic reconstruction and Poissonian failures is
Z(p) =
b

C(p) (C(p) +H)

+
(p)

(18)
where C(p) = C
0
+(340 [CU/m
3
] b h+5900 [CU/m
3
] a
s
) L [m] with C
0
= 300 and L the length
of the column. The cost H contains failure cost H
F
as well as cost H
M
for physical damage in case
of failure.
The acceptability criterion can be written as
5
p
C(p) G

`
kN
PE
5
p

+
(p)

(19)
Optimization will be performed for the public with benefit b = 0.25 and interest rate = 0.0175.
The LQI-data are the same as for the steel column example so that H
F
= 8.8 10
5
and G

`
kN
PE
=
1.9 10
6
for kN
PE
= 1.0. The cost H
M
for physical damage are set to H
M
= 0 and H
M
= 2.0 10
6
in two different optimization runs in order to express the influence on the optimal and acceptable
result.
The upper bound of the time variant failure probability in the stationary case is determined by
P
f
(p,t) = P
f
(p, t
1
) + (t
2
t
1
)
+
(p). The equivalent beta values are derived from the upper
bound solution. In the following table the optimal reinforcement area is calculated for various study
cases, where the column length, the cross section, the transversal distance a
1
and the longitudinal
distance a
2
to the surrounding columns and the number of storeys in the building is varied. A life
15
time (t
2
t
1
) = 50 years is assumed
Study Distance Cross Optim. equ. failure Optim. equ. failure
case section reinf. beta rate reinf. beta rate
area value area value
a
1
a
2
b h A
s
10
4
A
s
10
4
[m] [m] [mm] [m
2
]
h
1
50 years
i h
1
year
i
[m
2
]
h
1
50 years
i h
1
year
i
H
M
= 0 H
M
= 2.0 10
6
1 5 5 0.350.70 9.18 3.52 3.910
6
10.05 3.63 2.610
6
2 5 5 0.300.60 5.33 4.03 5.210
7
5.33 4.03 5.210
7
3 5 5 0.350.70 26.34 3.26 9.610
6
27.74 3.38 6.510
6
4 5 5 0.450.90 25.96 3.18 1.310
5
27.41 3.30 8.810
6
5 4 5 0.300.60 17.0 3.40 6.110
6
19.33 3.55 3.310
6
6 7 5 0.350.70 7.31 3.53 3.710
6
8.0 3.61 2.710
6
7 5 4 0.300.60 12.84 3.51 3.910
6
13.67 3.62 2.610
6
8 5 7 0.400.80 9.56 3.70 1.910
6
9.56 3.70 1.910
6
9 5 5 0.250.50 9.03 3.73 1.710
6
9.40 3.83 1.110
6
10 5 5 0.250.50 12.95 3.61 3.110
6
13.53 3.71 1.810
6
The length L [m] of the column is L
2
= 3, L
3
= 9, L
4
= 12 and L
i
= 6 in all other cases. The
number of storeys is 10, except for case 9 and 10 where it is 1 and 3, respectively.
In this example the LQI-criterion is active in all cases if the physical damage H
M
is not included.
If H
M
= 2.0 10
6
the optimal result is safer and therefore acceptable . The LQI-criterion is
not active. In case 2 and 8 the constraint on the minimal reinforcement area is dominant. The
resulting reliability indices per 50 years are within a narrowrange from3.2 to 4.0. The optimization
having regard to the acceptability criterion derived from the Life Quality Index produces a rather
uniform safety level in all study cases. The corresponding failure rates are comparatively low. As
pointed out in [3] the optimum failure rates achievable by a design according to EUROCODES
are by at least an order of magnitude smaller. Remember that kN
PE
= 1 has been assumed which
probably is unrealistically low. For this reason the effect of higher values of the number of fatalities
N
F
= kN
PE
in study case 1 on the failure rate is shown in figure 16. The LQI-data are the same
as before. The solid line is derived from eq. (19) and the dashed lines show the optimal results
determined fromeq. (18). The latter are calculated with the cost for physical damage H
M
= 2.010
6
and H
M
= 4.0 10
6
. The characteristics of the curves are the same as in the other study cases, if
no additional constraint is active. As it can be seen the influence of the cost for physical damage
diminishes with higer numbers of expected fatalities. The failure cost H
F
are dependent on N
F
whereas H
M
is a given value. Therefore, the ratio between H
F
and H
M
tends to H
F
with increasing
N
F
and the optimal curves coincide. In this example the optimal solution is only acceptable for
a small number of expected fatalities. With increasing N
F
the LQI-criterion becomes active. For
H
M
= 2.0 10
6
the intersection between the optimal and the limit curves occurs already at N
F
2.
The limit derived from the LQI-criterion has to be taken for N
F
& 2.5. A higer value of damage
cost in case of failure (H
M
= 4.0 10
6
) forces the optimal solution towards acceptable failure rates
until N
F
. 4.5.
7. Example 6: Slender steel column with combined loading and intermittencies
The next example is a steel column with I-section and flange breadth
b
, flange thickness
d
and
16
1 10 100 1
.
10
3 1
.
10
9
1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
f 0 := Log(N
F
)
L
o
g
(
f
a
i
l
u
r
e
r
a
t
e
)
H
M
= 2.010
6
H
M
= 4.010
6
1 10 100 1
.
10
3 1
.
10
9
1
.
10
8
1
.
10
7
1
.
10
6
1
.
10
5
f 0 := Log(N
F
)
L
o
g
(
f
a
i
l
u
r
e
r
a
t
e
)
H
M
= 2.010
6
H
M
= 4.010
6
Figure 1: Risk-Consequence curves for various cost of physical damage in study case 1. Dashed
lines correspond to optimal, the soild line to acceptable results.
height of steel profile
h
. The design parameter
d
and
h
will be changed in the cost optimization.
They are mean values of appropriate stochastic variables.
The objective function for systematic reconstruction and a stationary Poissonian renewal process
is again given by eq. (6). C(p) = C
0
+C
1
(2
b

d
+t
h
) s 8.2 10
9
are the construction cost of
the steel column with length s = 7500 [mm], initial cost C
0
= 100 [CU] and C
1
= 850 [CU/m
3
].
The stochastic characteristics of the random variables are given by
Stochastic variable Distr. Mean/St.deviat.
Yield stress F
S

N
mm
2

N 500/25
Load P
D
[N] N 400000/4000
Load P
1
[N] N 800000/150000
Load P
2
[N] LN 800000/150000
Load P
3
[N] LN 800000/200000
Web thickness t [mm] N 10/0.5
Flange Breadth B [mm] LN
b
= 300/0.1*
b
Flange Thickness D [mm] LN
d
/0.1*
d
Height of profile H [mm] LN
h
/0.1*
h
Initial deflection F
0
[mm] N m
F
0
/
F
0
Youngs Modulus E

N
mm
2

N 210000/4200
where the mean value m
F0
of F
0
is defined by
m
F
0
=
1
20
s
0.5 B D H
2
+
tH
3
12
2 B D +t H
+
s
500
and
F
0
= 0.3 m
F
0
. The limit state function in case of stability can be written approximately as
g(x, p) = F
S
P (
1
A
S
+
F
0
M
S


b

b
P
) (20)
17
where the auxiliary functions are defined by
P = P
D
+P
1
+P
2
+P
3
A
S
= 2BD (area of section)
M
S
= BDH (modulus of section)
M
i
=
1
2
BDH
2
(moment of inertia)

b
=

2
EM
i
s
2
(Euler buckling load)
Load P
1
is modelled as stationary differentiable Gaussian process with autocorrelation function

ij
() = exp(
2
). The loads P
2
and P
3
are rectangular wave renewal processes with jump rates

2
= 10.0 and
3
= 1.0 [1/year]. Also this mechanical model is an approximation which, however,
fully serves the purpose.
The acceptability criterion is given by
5
p
C(p) G

`
kN
PE
5
p
(
+
(p)) (21)
and the LQI-data are H
F
= 6.0 10
5
and G

`
kN
PE
= 1.9 10
6
(kN
PE
= 1). The direct cost are
H
M
= 3 C
0
= 300. The objective function Z(p) includes the cost H
F
for the public with b = 200
and = 0.0172.
The interarrival-duration intensities are given for the loads P
1
, P
2
and P
3
by
1
= 0.01,
2
= 1.0
and
3
= 100.0. The results for an optimization of the height
h
of the profile as design parameter
with given values
d
= 17.0 and
b
= 300 are
Optimal
h

i
: PD+ P1 P2 P3 P1 P1 P2 P1 +P2
+P2 +P3 +P3 +P3
348.6 13.6 7.7 7.1 6.2 5.3 4.94 4.93 3.6
Coincidence 4.9 4.9 4.9 4.9 4.9 4.9 4.9 4.9
Prob. 10
3
10
5
10
3
10
1
10
5
10
3
10
1
10
3

optimal
i
0 2.8 5.1 3.3 7.4 1.5 4.5 2.3
10
14
10
12
10
10
10
7
10
6
10
6
10
3
Coinc. Prob. 0 1.4 2.5 1.6 3.6 7.5 2.2 1.1

optimal
i
10
18
10
14
10
10
10
11
10
9
10
6
10
5
The LQI-criterion is active at the optimal solution.
Without including the LQI-criterion the results are
Optimal
h

i
: PD+ P1 P2 P3 P1 P1 P2 P1 +P2
+P2 +P3 +P3 +P3
346.1 13.5 7.7 7.1 6.1 5.3 4.91 4.90 3.5

optimal
i
0 4.0 6.7 3.9 9.1 1.8 5.3 2.6
10
14
10
12
10
10
10
7
10
6
10
6
10
3
Coinc. Prob. 0 2.0 3.3 1.9 4.5 8.8 2.6 1.3

optimal
i
10
18
10
14
10
10
10
11
10
9
10
6
10
5
8. Example 7: Reinforced concrete column under combined loading and intermitencies
This example is also based on [3] . The design parameter is the reinforcement area A
s
of the
square shape b h of the column cross section. A
s
should satisfy the condition 0.003 b h < a
s
<
0.08 b h. The limit state function is expressed as the difference between the resistance bending
18
moment M
R
and the actual bending moment M about the cross section centre
g(x, p) =
R
M
R

E
M (22)
where the model uncertainties
R
and
E
are normally distributed with mean 1.1 and 1.0 respec-
tively and coefficient of variation V = 0.1. Otherwise the same stochastic model for the strength
parametes as in the foregoing reinforcement concrete column example is chosen. The loads are
modelled as rectangular wave renewal processes with Gumbel distribution. It is distinguished
between wind pressure with mean 0.00035 [MN/m
2
], standard deviation 0.00006 and jump rate
= 10 [year], long term imposed load with mean 0.0006 [MN/m
2
], standard deviation corre-
sponding to the random point-in time distribution and jump rate = 1/7 [1/year] and short term
imposed load with mean 0.0002 [MN/m
2
], standard deviation corresponding to the distribution of
24 hours maximum and jump rate = n [1/year] (n = number of storeys).
The interarrival-duration intensities are given for the wind pressure P
W
, the long term imposed
load P
L
and the short term imposed load P
S
by
W
= 0.009,
L
= 3.0 and
S
= n/365.
An objective function for systematic reconstruction and Poissonian failures is given by
Z(p) =
b

C(p) (C(p) +H
F
)

+
(p)

(23)
where C(p) = C
0
+(125 [CU/m
3
] b h+4001 [CU/m
3
] A
s
) L [m] with C
0
= 300 and Lthe length
of the column.
The acceptability criterion can be written as
5
p
C(p) G

`
kN
PE
5
p

+
(p)

(24)
Optimization will be performed for the public with benefit b = 0.25 and interest rate = 0.0175.
The LQI-data are H
F
= 6.9 10
5
and G

`
kN
PE
= 1.9 10
6
with kN
PE
= 1.0. The upper
bound of the time variant failure probability in the stationary case is determined by P
f
(p,t) =
P
f
(p, t
1
) +(t
2
t
1
)
+
(p). The equivalent beta values are derived from the upper bound solution.
In the following table the optimal reinforcement area is calculated for various study cases, where
the column length, the cross section, the distance to the surrounding columns and the number of
storeys in the building is varied. A life time of 50 years is assumed.
Stu- Trans- Longi- Cross Optim. equ.
dy versal tudinal section reinf. beta
case dis dis- b h area value
tance tance [mm] A
s
10
4
[m] [m] [m
2
]
1 5 5 0.350.70 11.44 3.62
2 5 5 0.300.60 6.7 3.82
3 5 5 0.350.70 29.12 3.38
4 5 5 0.450.90 27.29 3.31
5 4 5 0.300.60 21.29 3.49
6 7 5 0.350.70 9.86 3.61
7 5 4 0.300.60 15.88 3.61
8 5 7 0.400.80 9.69 3.58
9 5 5 0.250.50 10.54 3.80
10 5 5 0.250.50 13.44 3.69
19
Study
i
: PW, PS PL PW PS PS PL PS
case PS, +PL +PW +PW +PL
PL = 0 +PW
1 9.86 9.85 9.34 3.91 9.27 3.92 4.00 4.01
Coinc. 2.4 7.2 6.5 2.2 2.0 6.5 5.9 1.8
Prob. 10
1
10
1
10
3
10
3
10
2
10
3
10
5
10
4

optimal
i
0 3.4 5.7 4.6 4.2 4.4 7.6 3.1
10
22
10
24
10
6
10
22
10
4
10
6
10
4
Coinc. Prob. 0 2.5 3.7 1.0 8.2 2.9 4.5 5.5

optimal
i
10
22
10
26
10
8
10
24
10
6
10
10
10
8
2 7.61 7.53 7.02 4.15 6.96 4.12 3.94 3.91
3 9.87 9.87 9.85 3.69 9.85 3.70 3.76 3.77

optimal
i
0 2.7 4.6 8.7 3.4 8.5 1.6 6.7
10
22
10
24
10
6
10
22
10
4
10
5
10
4
Coinc. Prob. 0 1.9 3.0 1.9 6.7 5.5 9.5 1.2

optimal
i
10
22
10
26
10
8
10
24
10
6
10
10
10
7
4 9.87 9.87 9.85 3.62 9.84 3.64 3.73 3.75
5 9.78 9.76 9.65 3.82 9.63 3.81 3.72 3.70
6 8.01 7.93 7.39 3.90 7.32 3.91 3.96 3.96
7 9.65 9.61 9.37 3.90 9.32 3.91 3.95 3.96
8 9.84 9.84 9.83 3.88 9.82 3.89 3.96 3.97
9 7.93 7.92 7.92 4.04 7.92 4.09 4.10 4.15
10 7.88 7.88 7.87 3.67 7.87 3.70 3.76 3.79
The length L [m] of the column is L
2
= 3, L
3
= 9, L
4
= 12 and L
i
= 6 in all other cases. The
number of storeys is 10, except for case 9 and 10 where it is 1 and 3, respectively.
In this example the LQI-criterion is active in all cases. The resulting reliability indices are within
a narrowrange from3.3 to 3.8. The optimization having regard to the acceptability criterion derived
from the Life Quality Index produces a rather uniform safety level in all study cases.
9. Example 8: Earthquake resistant design
A technically more involved example tries to establish a basis for the codification of design
values for earthquake resistant design. It introduces so-called risk integrals. We follow closely the
considerations in [8] but use slightly different data. In a seismic region Poissonian earthquakes
occur with rate = 2.9 [1/year]. Magnitudes between m
u
= 4.0 and m
o
= 7.5 are considered. A
truncated Weibull distribution (for maxima) has been found to model the data adequately
F
M
(m) =
exp

mom
mow

exp

momu
mow

1 exp

momu
m
o
w

k
(25)
with w = 4.35 and k = 8.11. These data are characteristic for an area with medium to high
seismicity. With the attenuation law
20
a = h(m, r) = b
1
exp(b
2
m)(r
2
+ 7.3
2
)
1/2
exp(b
3
r) = exp(b
2
m)b(r) (26)
where b
1
= 0.0955g, b
2
= 0.573, b
3
= 0.00587, one determines the density of peak ground
acceleration as
f
A
(a, r) =
k
(m
o
h
1
(a,r))
k1
(m
o
w)
k
exp

(m
o
h
1
(a,r))
m
o
w

dh
1
(a,r)
da
1 exp

momu
mow

k
(27)
with m
u
h
1
(a, r) =
1
b
2
ln(
a
b(r)
) m
o
. Possible epicentres are uniformly distributed around
the site in a radius of r
max
= 200 km. Hence, the density of peak ground acceleration is:
f
A
(a) =
Z
rmax
0
f
A
(a, r)
2r
r
2
max
dr (28)
Peak ground acceleration then varies with a coefficient of variation of V
A
= 1.55. The maximum
responses given peak ground acceleration vary log-normally with coefficient of variation of V
S
=
0.60. A simplified limit state function then is
g(X) = RKSAE 0 (29)
for shear resistance versus shear demand. Herein, R is a log-normal resistance with V
R
0.2, K
contains all system-specific properties and is, without loss of generality, assumed to equal unity, S
is the log-normal variability in the (elastic) spectral enhancement factor with mean m
S
= 1 and
A is peak ground acceleration. The systematic frequency-dependent part of S must be taken into
account in K. E is the log-normal error in relation (26) with mean 1 and coefficient of variation
V
E
0.6. The conditional failure probability (fragility curve) is
P
f
(p | a) =

ln
n
p
Km
S
am
E
q
(1+V
2
S
)(1+V
2
E
)
1+V
2
R
o
p
ln ((1 +V
2
R
)(1 +V
2
S
)(1 +V
2
E
))

(30)
with p = m
R
the design parameter because m
S
m
E
= 1.0. The objective function (without
benefit term) for systematic reconstruction is
Z(p) = C(p)f(N
PE
) + (31)
+E
A
"

C
R
(p, a)(1 P
f
(p | a))f(N
PE
)

+
+((C(p) +H
0
+H
M
(a))f(N
PE
) +H
F
(a))
P
f
(p|a)

#
The acceptability criterion eq. (??) to be used as a constraint for eq. (31) correspondingly reads:
d
dp

C
0
+C
1
p

E
A

`
1
2
(1 exp(0.25a))N
PE
d
dp
(P
f
(p, a))

(32)
The following widely verified relationship between peak ground acceleration and MSK-intensity
log(a) = 0.31MSK 2.5 is assumed. We distinguish between normal damage to the building
during an earthquake and building collapse. Construction, retrofitting, loss of business and physical
damage cost are slightly underproportional to the occupation rate of a unit in a residential building
so that we choose f(N
PE
) = (N
PE
/3)
0.8
. C(p) =

C
0
+C
1
p

is the construction cost, C

R
(p, a) =
21
C(p)(1 exp(0.25a)) is the cost of retrofitting taking account of the fact that retrofitting cost
approach the cost of complete reconstruction for larger a, H
M
(a) = H
M
a
0.4
is the physical damage
cost. The physical damage term includes infrastructure losses for large accelerations. Indirect
cost such as loss of business is approximated by H
0
= `C
0
. The estimation of human losses is
difficult. They also depend on a as people are increasingly trapped at collapse. Immediate death
then has probability 0.3 to 0.5 or more but some 10 to 20% die later in hospital. This leads to
H
F
(a) = SLSC
1
2
(1 exp(0.25a))N
PE
for the compensation cost for loss of human life. Note
that the factor
1
2
(1 exp(0.25a)) replaces the constant k in eq. (7). The constant 0.25 in these
relationships appears to vary with building type and material. The concavity of the function with
respect to a implies convexity with respect to MSK in agreement with estimates in the literature.
It is certainly only a rough approximation. Furthermore, we have C
0
= 10
6
, C
1
= 3 10
4
, = 1.1,
H
M
= 5 10
5
, = 0.02, N
PE
= 3. It is beyond the scope of this paper to discuss all these special
choices in detail but they are essentially in line with the findings in [1] and other sources in the
literature. The damage term in eq. (31) is conditional on a. The expectation operation removes the
condition. The damage term is also called risk integral. The table below collects typical data for
three different socio-economic levels.
Socio-economic level High Medium Low
GDP 23500 6500 1500
q 0.145 0.17 0.20
e 77 65 55
C

35 50 65
` 3.62 1.0 0.23
An additional FORM/SORM-analysis can then determine the design values a

corresponding to
p

and some other results of interest in the following results table.

Socio-economic level High Medium Low
p

4.80 4.00 3.54

h(p

) = P
f
(p

) (FORM) 3.710
4
6.310
4
8.810
4
Return period
1
h(p

)
2700 1600 1100
a

1.03 0.97 0.84

a

3.11 2.76 2.43

C(p

)/C
0
1.17 1.14 1.12
The design values of the accelerations a

have a return period of about 120 years. Roughly the

same design accelerations for all socio-economic levels indicate that the quantity E
A
[P
f
(p | a)]
decays very slowly with a. The values of a

< p

are also given. Surprisingly, the acceptance

criterion eq. (32) is not active and produces values p
lim
of approximately 1.5, 1.0 and 0.5, respec-
tively. These values imply an order of magnitude larger failure rates than the optimal solution.
This example is somewhat extreme because the loading side varies very much. Large changes in
the values of p

result in rather small changes in the failure rates. This explains why the differences
between the different socio-economic climates are relatively small. Also, in contrary to the previ-
ous examples, both construction cost and damage cost have been referred to a residential unit and
are roughly proportional to N
F
. Therefore, the effect of varying N
F
is insignificant and the failure
rates in the second line of the results table are the individual risks due to earthquakes in that region.
22
10. Conclusions
The examples show the application of the reliability-oriented cost-benefit optimization for struc-
tural components in the so-called one-level approach. An acceptability criterium as constraint of
the optimization problem ensures that the optimal structure is also a sufficiently safe structure.
Acceptibly safe structures are determined by the LQI-approach. Optimization is performed from
the publics and the owners point of view. The acceptability criterion becomes active especially if
the owner uses higher discount rates. The optimal results are significantly influenced by the sto-
chastic model chosen for the relevant uncertainties and the coefficient of variation of the resistance
variables and of the load processes.
Example 2 is proposed to be taken as standard as it can demonstrate the effect of many factors.
The few examples with realistic physical and stochastic models are well supporting the findings in
example 2. The computed acceptable risk-consequence curves show an almost perfect dependence
of the type r
acc
(N
F
) =
K
N
F
with K = 10
3
(case with large variabilities and/or costly safety
measures), 10
4
(standard case) and 10
5
(for small variabilities and inexpensive safety measures).
K has dimension [1/year]. Almost all examples with somehow realistic physical and/or stochastic
models and realistic parameters demonstrate that optimization leads generally to safer structures
and the constant K falls in between 10
4
and 10
5
for N
F
= 1.
References
[1] Coburn, A., Spence, R. (1992), Earthquake Protection, J. Wiley&Sons, Chichester
[2] Streicher, H., Rackwitz. R. (2004), Summary Report of DFG Research Project RA 322/9-2
NewConcepts for Load Combination, Institut fr Baustoffe und Konstruktion, TUMnchen
[3] Holick, M., Rackwitz, R. (1998) Time variant reliability of a column under variable loads
with intermittencies, Proc. European Conference on Safety and Reliability - ESREL98,
Trondheim, June 16-19 June, pp. 977-984.
[4] Rackwitz, R. (2000), Optimization - The Basis of Code Making and Reliability Verification,
Structural Safety, 22, 1, pp.27-60.
[5] Rackwitz, R. (2002), Optimization and Risk Acceptability based on the Life Quality Index,
Structural Safety, 24, pp. 297-331, 2002.
[6] Rackwitz, R. (2005), The Philosophy Behind the Life Quality Index and Empirical Verifica-
tion, Memorandum to JCSS, November 2005
[7] Rackwitz, R., Lentz, A., Faber, M. (2005), Socio-economically Sustainable Civil Engineer-
ing Infrastructures by Optimization, Structural Safety, 27, 3, 187-229
[8] Sanchez Silva, M., Rackwitz, R. (2004), Socio-economic Implications of the Life Quality
Index in the Design of OptimumStructures to Withstand Earthquakes, J. of Structures, ASCE
[9] Streicher, H., Rackwitz, R. (2004), Time-variant Reliability-oriented Structural Optimization
and a Renewal Model for Life-cycle Costing, Probabilistic Engineering Mechanics, 19, 1-2,
171-183
23